Error when performing inference on Triton Inference Server with embedded Stable Diffusion

When running the Stable Diffusion training model on a Triton Inference Server built on an on-premise server, the following error occurred.

Stub process is unhealthy

We have confirmed that the above error occurs during image inference by unet.

The server’s execution environment is as follows.
Note that Triton Inference Server is deployed in a Docker container.

However, the error no longer occurs when the device setting for executing the program is changed from GPU to CPU.

If anyone knows the cause, I would appreciate an answer.

The following part was incorrect

Container
– baseimage : nvcr.io/nvidia/tritonserver:22.05-py3

The correct content is as follows

Container
– baseimage : nvcr.io/nvidia/tritonserver:22.10-py3